Research map / Learning theory and algorithms / Kernels and regression theory
Optimal subsampling for high-dimensional partially linear models via machine learning methods
Yujing Shao, Lei Wang, Heng Lian, Haiying Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add A-/L-optimal Neyman-orthogonal subsampling probabilities instead of uniform (ablation) | Neyman-orthogonal subsample estimator · PLM case (b), q=200 | average CI length: 0.921 -> 0.752 (Lasso uniform vs A-opt, r=600) | Improved |
| use Neyman-orthogonal score with ML nuisance estimation instead of a linear model (ablation) | subsample estimator · PLM case (b), q=200 | coverage probability: 0.631 -> 0.953 (Linear vs Lasso A-opt, r=600) | Improved |
| use Neyman-orthogonal score with ML nuisance estimation instead of a linear model (ablation) | subsample estimator · PLM case (c), q=200 | coverage probability: 0.771 -> 0.954 (Linear vs Gbm A-opt, r=600) | Improved |
| use A-optimal rather than L-optimal probabilities | Gbm subsample estimator · PTS dataset (r=1000) | standard error of x1: 0.110 -> 0.113 (L-opt vs A-opt) | Mixed |
| use Gbm instead of Lasso as nuisance learner | A-optimal subsample estimator · PLM case (b), q=200 | coverage probability: 0.953 -> 0.962 (Lasso vs Gbm A-opt, r=600) | Improved |
| use Gbm instead of Rf as nuisance learner | subsample estimator · PLM case (b), n=5e5, p=200 | computation time (s): 48.50 -> 5.15 (T2, Rf vs Gbm) | Improved |
| subsample instead of using full data | Gbm subsample estimators · PTS dataset (r=1000) | computation time (s): 420.907 -> 8.897 | Improved |
| increase first-step proportion r0/(r0+r) from 0.1 toward 0.2, then beyond (ablation) | two-step algorithm · PLM case (b), q=200, r0+r=1500 | empirical MSE: approximately 0.2 for r0/(r0+r) | Mixed |
| increase subsample size r (ablation) | subsample estimators · PLM case (a), q=200 | average CI length: 0.613 -> 0.429 (oracle r=600 vs 1200) | Improved |
| use A-optimal rather than uniform sampling (ablation) | Gbm subsample estimator · PTS dataset (r=1000) | standard error of x1: 0.156 -> 0.113 (uniform vs A-opt) | Improved |
Closest papers by meaning
- Scale-invariant Optimal Sampling for Rare-events Data and Sparse Models (NeurIPS 2024 · similarity 0.84)
- Stochastic Optimization Algorithms for Instrumental Variable Regression with Streaming Data (NeurIPS 2024 · similarity 0.83)
- Knockoffs Inference for Partially Linear Models with Automatic Structure Discovery (AAAI 2025 · similarity 0.83)
- Optimal Nuisance Function Tuning for Estimating a Doubly Robust Functional under Proportional Asymptotics (NeurIPS 2025 · similarity 0.83)
- Approximating mutual information of high-dimensional variables using learned representations (NeurIPS 2024 · similarity 0.83)
- Nonparametric Instrumental Variable Regression through Stochastic Approximate Gradients (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
